Download Java and Python - Amazon Web Services

Java, Python, Zope and Indexing
Having Your Cake and Eating It
Chris Withers
[email protected]
http://www.zope.org/Members/chrisw
http://zope.nipltd.com/
Overview
• Java and Python Integration
• Indexing
– ZCatalog
– Lucene
http://zope.nipltd.com/
New Information Paradigms
(NIP)
• In Business 12 years
• Specialise in Knowledge & Content Management
• Customers include:
– Most large Pharmaceutical companies
– London Stock Exchange
– Readers Digest
http://zope.nipltd.com/
NIP’s Technologies
• Wide range of skills including:
– Zope Consulting & Hosting
– J2EE and Oracle
– Lotus Notes
• Operating Systems:
– Windows
– Solaris
– Linux
http://zope.nipltd.com/
Contacting NIP
• http://www.nipltd.com
– For an overview
• http://zope.nipltd.com
– For Zope specific stuff
• [email protected]
– To contact by email
http://zope.nipltd.com/
Java and Python
• Why use Java?
–
–
–
–
It’s overly verbose
Not very dynamic
Painful Exception Handling
“Too” object oriented
• But…
http://zope.nipltd.com/
Java and Python
• Why use Java?
– Quicker Execution
– Very Popular Language
• More libraries
• More robust
– more testers
• Better documentation
– more authors around
– Politically acceptable
http://zope.nipltd.com/
But I want to use Python!
…so find a way to use Java and Python in the same
environment.
• So what are the options?
– Jython / JPython
– Web Services
…and other loose couplings
– Java Python Environment
http://zope.nipltd.com/
Jython / JPython
• Python implemented in Java instead of C
+ Very politically acceptable
– Can’t use C extensions to Python
– Not the “main branch” of Python development
• Status?
http://zope.nipltd.com/
Loose Couplings
• Web Services
• Shared Files
• Low-level socket protocols
+ No restrictions on versions of languages or extensions
to languages used.
+ Easy to distribute applications over several machines
– A lot more work for the developer
– Inefficient communication between virtual machines
http://zope.nipltd.com/
Java Python Environment (JPE)
• Low-level bridge between a Java virtual machine
and a Python virtual machine
+ Use almost any Java library from Python
+ Use almost any Python library from Java
+ Very Transparent
– Difficult to Build, Install and find out about
http://zope.nipltd.com/
How does JPE work?
Java
Virtual Machine
JPE
Python
Virtual Machine
• Bridge written mainly in Python and Java
• C extension to Python (wrapped in Python package)
• C extension to Java (wrapped in Java package)
http://zope.nipltd.com/
So lets see it in action…
• Using Java from Python
• Using Python from Java
http://zope.nipltd.com/
Using Java from Python
import java
if not java.isInitialized():
java.initialize()
out = java.importClass( 'java.lang.System').out
out.println('Hello Python World from Java')
• How about a demo?
http://zope.nipltd.com/
Using Python from Java
import python.PyModule;
import python.PyObject;
class HelloWorld
{
static void main( String args[])
{
PyModule sys = new PyModule( "sys");
PyObject stdout = (PyObject)sys.getattr(
"stdout");
stdout.callmethod( "write", new PyObject[]
{ PyObject.asPython( "Hello Java world from
Python\n")});
}
}
• How about a demo?
http://zope.nipltd.com/
What are the problems?
• Needs the environment correctly set up
– Python & Java versions important
– PATH, CLASSPATH & PYTHONPATH important
• Difficult to build
– See How-To
– DON’T use nmake install!
• Performance
– But only in recent versions!
http://zope.nipltd.com/
Questions ?
http://zope.nipltd.com/
Indexing
• What do we mean by indexing?
–
–
–
–
Numbers
Dates
Text
Sorting in Relevance Ranking
• It’s a HARD problem!
– Don’t let Google fool you…
http://zope.nipltd.com/
What are the options?
• Commercial Solutions
– Verity
– Google boxes
– $$$ 
• ZCatalog
• Lucene
http://zope.nipltd.com/
ZCatalog
• Solves generic indexing problem for Zope
• Stores information in ZODB
– Participates in transaction framework 
– Stores all old revisions 
• TextIndex has very limited functionality
http://zope.nipltd.com/
Lucene
• Written originally by Doug Cutting
– Xerox's Palo Alto Research Center (PARC)
– Apple
– Excite@Home
•
•
•
•
Now part of the Apache Jakarta project
Only tackles text indexing
High Perforance
Fully Featured
– Phrase matching
• Written in Java 
http://zope.nipltd.com/
Let’s see some code…
…written in Python!
• Indexing Files
• Searching Indexed Files
http://zope.nipltd.com/
How does Lucene handle concurrency?
• File locks
• Never add to an existing index
Reader
1
2
Optimize
http://zope.nipltd.com/
Writer
3
4
1
2
3
LuceneIndex
• A PluggableIndex for Zope 2’s ZCatalog
• Really painful to implement 
– PluggableIndexes are Clunky
– Undocumented reliance on id attribute
– Really hoping it’ll be better in Zope 3…
• Lets have a look…
http://zope.nipltd.com/
A Comparison
• 1000 Documents, Average length 5781 Bytes
700
Indexing
600
Peripheral Stuff
Time (seconds)
500
400
300
200
100
0
Lucene through Java
http://zope.nipltd.com/
TextIndex
Lucene through JPE
LuceneIndex
Was that fair?
• Performance for BIG numbers of long documents
• TextIndex doesn’t do phrase matching
– Something which did took MUCH longer
• Lucene doesn’t support undo
– Do we care?
• LuceneIndex proved that the Lucene architecture
and ZCatalog’s architecture aren’t very compatible

http://zope.nipltd.com/
Conclusions
You can have your cake and eat it…
…just slowly 
…for now 
http://zope.nipltd.com/
Where from here?
• Optimise JPE?
• CORBA?
• Re-implement Lucene in Python?
• TextIndexNG?
http://zope.nipltd.com/
Questions ?
http://zope.nipltd.com/
Thankyou!
(PS: Swishdot is still on the way ;-)
(PPS: It was Steve A’s birthday yesterday!)
http://zope.nipltd.com/